The prompt
Copy it whole, replace every bracketed variable, then paste it as your first message.
You are my drafting partner for an essay. I direct, you propose, I judge, I own the result. Your first job is to interrogate, not to draft. Read everything before replying.
## 1. The essay
- The claim, in my words: [THE CLAIM]
- Why I can make it: [WHY YOU]
- The doubt I hold about it: [THE DOUBT]
- Reader and venue: [READER AND VENUE]
## 2. Raw material (the only source of scenes, facts, quotes, and sources)
[RAW MATERIAL]
Fact and material rules:
- Every scene, detail, number, quote, and source in the essay comes from the raw material or from my answers in this conversation.
- Never invent a memory, a line someone said, a study, a statistic, an expert, or a book. Where the essay wants one, write [VERIFY: what is needed] for facts and [GAP: what scene or detail is needed] for lived material, and continue.
- Do not add "both sides", a counterargument, or a concession unless I supplied it or asked for it. My doubt in section 1 is the counterargument; use that.
- Sources I name may be cited only for what the raw material says they say.
## 3. Voice
[VOICE PROFILE]
Banned (never use, even in options):
- [BANNED LIST]
- "In this essay", "I want to explore", "This got me thinking"
- Opening with a dictionary definition, a famous quote, or a rhetorical question
- "I learned that", "Looking back", "At the end of the day", "the journey"
- A final paragraph that states the lesson or restates the claim
- Universalizing the ending ("We all...", "Perhaps everyone...")
- Three-item lists used for rhythm
## 4. Interrogation (do this first)
Before any drafting, ask me five questions, one at a time if the interface allows, otherwise numbered. The questions must do these jobs:
1. Ask for a specific scene the raw material mentions but does not show: when, where, what was said, what I did with my hands.
2. Ask what I believed about this before, and the moment it changed, if there was one.
3. Push on the doubt: ask what would have to be true for the doubt to be right.
4. Ask which detail in the raw material I would be most embarrassed to leave out, and which I would be most embarrassed to include.
5. Ask what I want the reader to do, think, or feel differently, in one line.
Do not comment on my answers. Do not praise them.
## 5. Drafting rules (after the interrogation)
When I say "draft":
- One draft, to the length in section 1, then stop. No commentary.
- Open inside a scene or a specific fact, not with the claim.
- The claim appears in plain words once, where it has been earned, usually past the midpoint.
- The doubt appears and is taken seriously; it may be answered, or left standing. It may not be dismissed in a sentence.
- Use my answers from the interrogation verbatim where they are good. Mark them with nothing; just use them.
- End on a scene, an image, or a fact. Not on the claim, and not on a lesson.
Turn rules:
- "change only X": return only the changed sentences marked OLD and NEW.
- "options for X": three versions of X only, clearly different, labeled by tradeoff.
- "mine": the pasted text is my writing; do not revise it unless I name the issue.
- Never say you have verified anything. Mark uncertainty with [VERIFY].
## 6. Begin
Reply with one sentence restating the claim in your words, then the first interrogation question. Nothing else.- Best for
- Personal essays built on experience you actually had
- Argumentative essays and op-eds where you hold a position
- Longform newsletter pieces that need an argument, not just an observation
- Writers who know what they think but keep producing a tidy, evasive draft
- Not for
- Academic submissions where AI drafting is prohibited. Use only the interrogation phase, and check your institution's policy first
- Essays on topics you have no experience or research in. The prompt will return [VERIFY] and [GAP] tags where the essay should be
- Applications and admissions essays. Those have specific norms and a separate prompt in this library handles cover letters
An essay is a person thinking in public. The model cannot think in public for you. It can only arrange plausible sentences about a topic. That is why AI-assisted essays so often read as balanced, reasonable, and empty: they have a topic where the claim should be, and general observations where the life should be.
This prompt pushes the other way. Before anything is drafted, the assistant works as an interrogator: it asks for the scene you have not told, the moment you changed your mind, the evidence you are leaning on hardest. It is forbidden from supplying 'both sides' unless you asked for them, and from inventing the specifics that make an essay feel lived. When it drafts, it drafts from your answers.
Students should read the ethics notes before using this. Many institutions treat AI-drafted essays as misconduct regardless of how much you revise. The interrogation half of this prompt is useful on its own, and is usually permitted, but the drafting half may not be.
Variables
What to put in each bracket. Required ones are marked. Thin answers here are the usual cause of generic output.
[THE CLAIM]RequiredWhat you are arguing, in one sentence you would say out loud to a friend. Not the topic, the claim.
ExampleLearning to cook badly in my thirties did more for my marriage than any conversation we had about chores
[WHY YOU]RequiredWhat you lived, saw, read, or did that gives you standing to make the claim.
ExampleEight years of being the one who did not cook, then a year of cooking three nights a week, badly, and the specific things that changed
[RAW MATERIAL]RequiredScenes, moments, lines people said, numbers, notes, sources you have actually read. Messy is fine. This is the only material the essay may use.
ExampleThe night of the burned rice; what my partner said about the smoke alarm; the chore spreadsheet we abandoned; a passage from a book on household labor with the page number
[THE DOUBT]RequiredThe strongest objection to your claim, or the thing you are not sure about. The essay will be stronger for facing it.
ExampleMaybe it was not the cooking; maybe it was that I finally did anything at all
[READER AND VENUE]RequiredWho will read it and where.
ExampleA general-interest online magazine's readers, mostly people in long relationships, 1,500 words
[VOICE PROFILE]RequiredYour Voice Capture profile or a sample paragraph of your own nonfiction.
ExamplePaste a paragraph from an essay you have written
[BANNED LIST]Moves you do not want, beyond the defaults.
ExampleThe word 'journey', 'I learned that', any sentence starting with 'Ultimately'
Why it is built this way
Each block does one job. Knowing the job lets you shorten or adapt the prompt without breaking it.
Interrogation before drafting
The most valuable thing an assistant can do for an essay is ask the questions a good editor asks. The five jobs are specific so the questions are not generic 'tell me more' prompts.
The claim and the doubt as separate variables
An essay with no claim is a topic. An essay with no doubt is a lecture. Making both explicit keeps the model from supplying the default 'balanced' structure and makes your actual uncertainty the counterargument.
[GAP] for lived material, [VERIFY] for facts
Two different kinds of hole. A fact can be looked up, but a scene can only be remembered by you. Separating them tells you what kind of work each gap needs.
No both-sides unless supplied
Models add concessions reflexively, and the result is an essay that argues nothing. Your doubt is the real objection, and the prompt routes the model to that instead.
The banned essay moves
Dictionary openers, 'I learned that', the universalizing close. Each is a shape the model reaches for because essays it has seen reach for them. Named bans work, but tone instructions do not.
Claim past the midpoint, ending on a scene
The essay shape that earns the claim rather than announcing it. Stated as a rule because the model's default is thesis first, lesson last.
Example run
A representative first response, with the input summarized. Your results will differ by model and material.
Filled for the cooking-badly essay: the claim about the marriage, eight years of not cooking, the burned rice and smoke alarm night, the abandoned chore spreadsheet, a cited passage on household labor, and the doubt that it was simply doing anything at all.
Claim as I understand it: cooking badly, and regularly, changed your marriage more than any explicit negotiation about chores did, and the badness was part of why. First question. The raw material mentions "the night of the burned rice" and the smoke alarm, but does not show it. Where were you both standing when the alarm went off, what did your partner say first, and what did you do with the pan?
The first question went straight to the scene the essay needed, and the writer's answer (she put the pan in the sink and her partner opened a window and said nothing, then laughed) became the opening paragraph nearly unchanged. The fourth question, about the detail she would be embarrassed to include, produced the essay's honest center: that she had enjoyed being bad at something in front of someone.
Follow-up prompts
The second turn is where vibe writing happens. These are the short messages that keep the session moving.
After the five questions, to get the draft
Draft.
When the draft states the claim too early
The claim appears in paragraph [NUMBER]. Move it to after the [SCENE] and change only what is needed to make that work. OLD and NEW.
When the doubt got dismissed
The doubt is handled in one sentence. Change only that paragraph: take the doubt seriously for at least four sentences, using what I said in my answer to question 3, and do not resolve it.
When a scene has detail you did not supply
List every concrete detail in the draft and where it came from: raw material, my answers, or neither. I will cut the 'neither' ones.
When you want to write the ending yourself
Do not draft an ending. Give me the last image from my raw material that the essay has not yet used, and ask me one question about it.
When it goes wrong
Symptoms you will actually see, the usual cause, and the correction that works.
The essay is balanced and reasonable and says nothing.
- Cause
- The claim variable was a topic ('the division of household labor') rather than a claim.
- Fix
- Rewrite the claim as something a friend could disagree with. Re-run the interrogation. The questions will be sharper.
Memories appear that you do not have.
- Cause
- The model filled a [GAP] with a plausible scene because the structure needed one.
- Fix
- Run the detail-source follow-up and cut everything marked 'neither'. This is not a style problem: a fabricated memory in a personal essay is a lie, even a small one.
The ending universalizes: 'Perhaps all of us...'
- Cause
- The strongest essay reflex in the model, and one that survives the ban in some models.
- Fix
- Cut the last paragraph. Use the 'write the ending yourself' follow-up and finish it by hand.
The interrogation questions are generic.
- Cause
- Thin raw material. The questions can only be as specific as what they are asking about.
- Fix
- Add more raw material, especially fragments: a line someone said, a thing you noticed. Then say: 'Ask again, and each question must refer to something specific in the material.'
Before you publish
The prompt cannot do these. You can.
- Every scene in the essay happened as described. Check each against memory, and where memory is uncertain, say so in the essay. Readers trust that.
- Every cited source says what the essay says it says. Open it. Quote accurately.
- The claim is one you would defend to someone who read the essay and disagreed.
- If this is for a class or an institution with an AI policy, follow it, including disclosure. If you used only the interrogation, that is usually fine to say. If you used the draft, say that.
- Read it aloud. The sentences that embarrass you slightly are often the true ones. The sentences that sound wise are often the model's.
Questions
Is it okay to use this for a college essay?
Check the policy first. It varies by institution and by course, and many prohibit AI-generated text outright. The interrogation phase, where the assistant only asks questions, is permitted in many places and is the most useful half. If drafting is prohibited, stop after the questions and write from your answers.
What if I do not have a doubt about my claim?
Then you have not thought about it hard enough, or the claim is too safe. Ask the assistant for the strongest objection a smart reader would raise, pick the one that worries you, and put that in the variable. The essay will be better for it.
How is this different from the Blog Post prompt?
A blog post serves a search query and an angle. An essay serves a claim and the thinking behind it. The blog prompt structures around headers and intent. This one structures around interrogation, scene, and doubt, and has no search component at all.
Can the assistant suggest sources?
It can suggest where to look, but anything it names must be treated as a lead, not a source. The prompt forbids citing anything not in your raw material for exactly this reason. Find the source yourself, read it, then add it.